Equivalent GP models, different results

Questions v5
, ,
3

In both cases I’m calling it just after calling sample_prior_predictive. I wrapped them both in a function:

idata = az.InferenceData()
with model_train:
     idata.extend(sample_prior_predictive(**kwargs_prior), join='right')
     idata.extend(sample(**kwargs_posterior), join='right')

Could it be that deepcopy is making problems by not copying the tensors properly?